What are the most commonly reverted file types?

According to the skill's benchmarks, Emacs org-mode files (.org), Python scripts (.py), and Shell scripts (.sh) are the most frequently reverted file types.

Can I query my own thread history with this skill?

Yes, the skill uses a DuckDB database to index local Amp file-changes, allowing you to run SQL queries to find specific patterns or revert percentages across your history.

Does this skill require an external API?

Currently, Amp-Skill operates locally by processing file-change diffs stored in your local .amp directory, though it has potential for future GraphQL API integration.

How does this skill help improve agent behavior?

By identifying high-rejection threads and specific file-type sensitivities (like .org or .sh files), developers can implement confirmation prompts or alternative logic to avoid common failure points.

What is a 'two-lock cascade' in Amp-Skill?

A two-lock cascade occurs when an agent attempts the same action multiple times consecutively within a short period (e.g., 60 seconds) despite the user rejecting or reverting those actions.

Amp Interruption Analytics

Name: Amp Interruption Analytics
Author: plurigrid

byplurigrid

•

分析と監視

Analyzes tool rejection patterns and interruption history to improve AI agent reliability and behavior.

Amp-Skill provides deep insights into how AI agents are interrupted during tasks by distilling usage patterns from Amp thread history. Built on a DuckDB foundation, it identifies critical rejection scenarios such as 'two-lock' cascades—where an agent repeatedly attempts rejected actions—and high-rejection threads involving sensitive file types like .org or security scripts. By analyzing abandonment and revert data, this skill enables developers to implement smarter backoff strategies and coordinate alternative approaches using triadic logic, ultimately leading to more responsive and context-aware agent interactions.

主な機能

01File-type sensitivity analysis to highlight high-risk or frequently rejected operations

022 GitHub stars

03SQL-powered querying of thread history and tool call metrics via DuckDB

04GF(3) triadic role assignment for coordinating coordination and generation layers

05Detection of five interruption patterns: abandoned, reverted, two_lock, discontinuation, and high_rejection

06Identification of two-lock cascades to prevent repetitive AI behavior loops

ユースケース

01Implementing security-conscious backoff strategies for sensitive file edits

02Analyzing user abandonment patterns to refine thread completion rates

03Debugging repetitive tool-call loops where the agent ignores user rejections

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add plurigrid/asi amp-skill

For use in Claude.ai and ChatGPT